samcheng0/deeplm-108m

Deeplm — Auto-upload (step 2000)

Model checkpoint otomatis dari training loop. Setiap 100 langkah, seluruh state dikirim ke HF.

Training Metrics

• step: 2000
• loss: 9.9303
• lr: 0.0001
• grad_norm: 1.0000
• phase: exploration
• state: init
• confidence: 0

Charts

Included Files (setiap upload)

File	Deskripsi
model.safetensors	BitNet ternary weights
config.json	Model configuration
tokenizer.json	BPETokenizer
tokenizer_config.json	Tokenizer config
checkpoint-2000/	Full checkpoint (model.pt, optimizer.pt, training_state.json)
charts/	Training visualization PNGs
metrics.jsonl	Full training log (all steps)
tuner_state.json	AutoTuner internal state
README.md	This file

Penggunaan

import sys; sys.path.insert(0, 'deeplm')
from deeplm.config import DeeplmConfig
from deeplm.model.deeplm import DeeplmModel
from safetensors.torch import load_file

config = DeeplmConfig()
model = DeeplmModel(config)
state_dict = load_file('model.safetensors')
model.load_state_dict(state_dict, strict=False)